word substitution perturbation

deep nlp model (1) randomized perturbation defense (1) robustness enhancement (1) nlp model (1) word perturbation (1) word substitution (1) adversarial attack (1) deep learning (1) model robustness (1) adversarial robustness (1)